Illegal and artisanal small-scale mining detection

Freshly disturbed laterite and turbid river plumes leave unmistakable spectral fingerprints in open satellite archives. Sentinel-2, Landsat and Planet imagery, read together, can date the onset of illegal ASM activity and track its spread week by week.

Sensors

Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 20 m in red-edge and SWIR. Five-day revisit at the equator (two satellites combined). SWIR bands (B11 at 1610 nm, B12 at 2190 nm) are particularly sensitive to bare soil and moist sediment exposure. Free archive from 2015 provides the temporal baseline needed to date disturbance onset.
Landsat 8/9 OLI: 30 m multispectral resolution, 16-day single-satellite revisit (8-day combined for 8 and 9). OLI's coastal-aerosol band (Band 1, 443 nm) and SWIR bands support turbidity mapping in rivers and sediment-exposure indices. Archive extends to 1972 across the Landsat series, giving unmatched historical context for long-running sites.
Planet SuperDove: 3 m resolution, near-daily global revisit. Eight spectral bands including red-edge and two NIR channels. Used to pin the active working face within a known disturbance polygon and to detect new clearings too small (under 0.1 ha) to register clearly in Sentinel-2. Commercial tasking required; latency as low as same-day.
Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral. Single-pass stereo capable. Used for confirmation imagery when a regulatory authority needs court-admissible evidence of equipment, access tracks or pit morphology at a specific site. Not a monitoring tool at scale; tasked on demand once lower-resolution analysis has flagged a priority location.

What freshly stripped laterite looks like from orbit

Artisanal and small-scale mining in tropical regions almost always begins with vegetation clearance and soil stripping. Laterite, the iron-rich weathered regolith common across West Africa, the Amazon basin and Southeast Asia, has a distinctive high reflectance in the shortwave infrared and a suppressed vegetation signal in the near-infrared. The contrast with surrounding forest canopy is sharp enough that a single Sentinel-2 scene can reveal clearings as small as 0.1 ha, well within the footprint of a hand-dug pit cluster.

The Bare Soil Index (BSI) and the Iron Oxide Ratio (OLI Band 4 / Band 2, or equivalent Sentinel-2 B4/B2) are the two most commonly applied indices in published ASM detection studies. Neither is exotic; both are derived from standard multispectral bands available in every Sentinel-2 and Landsat 9 scene. The real analytical work is temporal: a single high-BSI pixel proves nothing. A pixel that transitions from a low vegetation index to a high bare-soil index between two consecutive acquisitions is a disturbance event with a date attached.

Rivers as unwilling informants

ASM operations that work alluvial deposits, particularly gold dredging on Amazonian and West African rivers, inject fine sediment into the water column continuously. That suspended sediment raises water-leaving reflectance in the visible and red bands in a way that is measurable in Sentinel-2 and Landsat data. The Normalised Difference Turbidity Index (NDTI), using red and green bands, produces a proxy for total suspended solids that tracks well against in-situ measurements in published literature, though the relationship is site-specific and benefits from local calibration.

Turbidity plumes are useful precisely because they are hard to conceal. A miner can cover a pit with a tarpaulin. They cannot stop a river carrying their sediment downstream. Plume analysis also extends the spatial reach of detection: a mining operation in a dense forest gap that is too small or too cloud-obscured to detect directly may still produce a detectable signal 5 to 20 km downstream. The honest caveat is that natural flood events and agricultural runoff produce similar signals, so plume detections require corroboration from the land-surface record before they become actionable.

Dating onset and tracking spread

The temporal archive is where open data earns its keep. Sentinel-2's five-day revisit (two-satellite combined) means that in cloud-free conditions a new clearing can be dated to within a week of its appearance. In practice, persistent cloud cover over tropical regions is the binding constraint. The Amazon basin can be overcast for weeks at a time; West Africa's harmattan season is the exception rather than the rule. Analysts working these regions typically construct dense time-series composites, accepting some temporal blur in exchange for cloud-free coverage.

Landsat's 16-day revisit is coarser, but its archive back to the early 1970s allows analysts to establish whether a site has a history of activity. Many nominally new ASM detections turn out to be re-activations of sites that operated during previous commodity price spikes. That historical context matters for enforcement: a site with a 20-year activity record is a different regulatory problem from a site that opened last month.

Planet SuperDove resolves the temporal ambiguity at the cost of spectral depth. Its near-daily 3 m imagery can show the working face moving across a site day by day, which is useful for estimating the rate of land disturbance and, indirectly, the scale of operation. It does not replace the multispectral archive; it narrows the uncertainty that the archive leaves behind.

Arid zones: a different signature, the same physics

Coverage of ASM detection tends to focus on tropical forest regions because the contrast between vegetation and bare soil is so stark. But artisanal mining in arid and semi-arid zones, common across the Sahel, parts of Central Asia and the Atacama fringe, presents a different problem. There is no forest canopy to breach, so the disturbance signature is subtler: fresh excavation material has a different mineralogical composition and moisture content from the surrounding surface, producing small but detectable differences in SWIR reflectance and surface temperature.

In these environments, shadow geometry from spoil heaps and the linear geometry of access tracks often provide clearer detection cues than spectral indices alone. Object-based image analysis, which groups pixels by shape and texture rather than treating each independently, is better suited here than simple index thresholding. Resolution matters more too: at 30 m, a small arid-zone pit may occupy only a handful of pixels, making Planet or Pléiades the operationally relevant sensors for initial detection, with Sentinel-2 and Landsat used to establish the temporal record once a site is known.

Limits that buyers need to understand

No satellite-based ASM detection system is a complete enforcement solution. Cloud cover is the most obvious limit, but not the only one. Optical sensors cannot see through the forest canopy to detect underground workings or tunnel-based alluvial extraction. Very small operations, a single family working a 20 m² pit, may fall below the detection threshold of Sentinel-2 and Landsat entirely, and even Planet's 3 m imagery may not produce a confident detection if the disturbed area is smaller than a few pixels.

Spectral confusion with legal activities is real. Smallholder agriculture, road construction and legal quarrying can produce bare-soil signatures indistinguishable from ASM at the pixel level. Contextual rules, proximity to known mineralised zones, absence of planning records, characteristic pit morphology, are necessary to reduce false positives to an acceptable rate. Satellize applies these contextual filters as part of its analytics pipeline; the Tonga crop-estimation programme demonstrated the same principle of combining spectral outputs with local ground-truth logic to improve classification reliability.

Revisit and latency depend on the sensor tier. Sentinel-2 and Landsat data are free and arrive within hours of acquisition, but cloud screening and compositing add processing time. Planet and Pléiades tasking introduces commercial cost and scheduling lead time. For operational monitoring programmes, a tiered architecture, open data for wide-area surveillance, commercial tasking for priority sites, is both technically and economically sensible.

From detection to evidence

Regulatory agencies and mining ministries increasingly need outputs that can survive legal scrutiny. A pixel classification map is not evidence; a time-stamped sequence of georeferenced imagery showing progressive disturbance, with metadata tracing the acquisition to a named satellite and a verified overpass time, is considerably closer. Pléiades Neo at 30 cm can resolve individual excavators and sluice boxes, which moves the output from 'anomaly flagged' to 'activity confirmed'.

The practical workflow runs in one direction: open-archive surveillance flags candidate sites at scale, commercial tasking confirms the highest-priority cases, and the resulting dossier, imagery, indices, change dates and coordinates, is formatted for the relevant authority. That last step, translating satellite output into a form that a mining inspector or prosecutor can act on, is where most remote-sensing projects stall. It is worth specifying deliverable format before the analysis begins, not after.

Typical figures

Spatial resolution (surveillance tier)	10 m (Sentinel-2 visible/NIR), 30 m (Landsat 8/9 OLI)
Spatial resolution (confirmation tier)	3 m (Planet SuperDove), 30 cm pan / 1.2 m multispectral (Pléiades Neo)
Revisit (surveillance tier)	5 days combined (Sentinel-2A+2B); 8 days combined (Landsat 8+9)
Revisit (confirmation tier)	Near-daily (Planet); on-demand tasking (Pléiades Neo)
Minimum detectable disturbance	~0.1 ha in Sentinel-2 (single-scene); ~0.01 ha indicative in Planet 3 m imagery; smaller pits require Pléiades
Key spectral bands	SWIR (1610 nm, 2190 nm), NIR (~865 nm), Red (~665 nm), Green (~560 nm) for BSI, Iron Oxide Ratio and NDTI
Archive depth	Sentinel-2 from 2015; Landsat from 1972 (Landsat 1 MSS); Planet from ~2016
Data latency (open archive)	Sentinel-2 and Landsat scenes available within 3–6 hours of acquisition via Copernicus Data Space and USGS EarthExplorer
Primary cloud constraint	Optical sensors only; persistent cloud cover in humid tropics can reduce effective revisit to 20–40 days in worst seasons
Typical delivery formats	GeoTIFF change-detection layers, GeoPackage / Shapefile site polygons, PDF site-evidence dossiers, GIS-ready alert feeds (GeoJSON)

Analytics Satellize can run

Wide-area disturbance alert layer	Bi-temporal and time-series BSI and NDVI change detection on Sentinel-2 and Landsat 8/9 composites; threshold-based flagging of new bare-soil exposures	Monthly GeoTIFF and GeoJSON alert layer covering a defined region of interest, with disturbance onset date per polygon
River turbidity anomaly map	NDTI time-series on Sentinel-2 red and green bands; anomaly scored against seasonal baseline per river reach	Bi-weekly turbidity anomaly raster and flagged reach shapefile, with upstream source-area candidates identified
Site activity timeline	Dense time-series compositing of all available Sentinel-2 and Planet acquisitions over a confirmed site; change-point detection to identify active and dormant periods	Per-site PDF dossier with annotated image sequence, disturbance area estimates and activity timeline chart
Active working-face delineation	Object-based image analysis on Planet SuperDove 3 m imagery; texture and spectral segmentation to separate fresh excavation from older spoil and access tracks	GeoPackage polygon layer of active working faces, updated on available Planet acquisitions (typically weekly)
Historical site inventory	Landsat archive analysis (1972 to present) using Iron Oxide Ratio and BSI to identify prior disturbance episodes at current ASM locations	Spreadsheet and shapefile inventory of confirmed sites with first-detection date, peak extent and activity episode count
High-resolution confirmation imagery package	On-demand Pléiades Neo tasking over priority sites identified in surveillance tier; pan-sharpened 30 cm product with georeferencing metadata	Georeferenced GeoTIFF with acquisition certificate, suitable for regulatory or legal submission
Contextual false-positive filter	Overlay of disturbance detections against licensed concession boundaries, road infrastructure buffers and agricultural land-use maps; rule-based exclusion of likely non-ASM sources	Filtered alert layer with confidence classification (high / medium / requires field verification) and exclusion-reason annotation

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.